Upload
vubao
View
218
Download
0
Embed Size (px)
Citation preview
Fall 2014—Honor Thesis--Statistics
Joint Model for Exchange Rate Dynamics and
Influence of Presetting Correlation between Stock
Price and Exchange Market
Jian Wang
University of California, Berkeley
I. INTRODUCTION:
American Depository Receipts (ADRs) are securities issued by non-US companies to cross list
their stocks on both domestic and foreign stock exchanges. In addition to stocks, there are also
options traded in these markets denominated in their respective currencies. As such, ADR
options are exposed to foreign exchange rate risk. In the summer research in Columbia
University we studied the implied volatilities of options issued by a non-US company and their
US traded ADR counterparts. Under the stochastic assumption of exchange rate we applied the
joint model for implied volatility and exchange rate to explain the options price discrepancies.
For each company by mean square minimization we extracted the exchange rate implied
volatility, which turned to be pretty effective and could catch the clue of currency depreciation
up to three months in advance. In the dynamic model, the correlation between stock price and
exchange rate was a critical factor determining the exchange rate implied volatility and we
maximized the mean square error over the exchange rate volatility and correlation. Yet the
implied correlation was quite unstable and if we constrained the correlation between -1 and 1 we
actually got lots of extreme implied correlation, which was not realistic. In the independent
research this semester, we can examine the sensitivity of implied exchange rate volatility respect
to correlation. By reasonable preset of the correlation, the expression of implied volatility
becomes explicit for direct analysis. There are three different measurements of correlation,
dynamic correlation over a short period, correlation as regression line over long period and
industrialized correlation to exchange rate. Though different measurement of correlation between
foreign stock market and exchange rate caught the similar trend, they gave significant different
levels of implied exchange rate volatility. By examining the results of three measurements of
correlation, we can compare the advantages and limitations for these methods. The study
improves the model we built in the summer to extract foreign exchange rate implied volatility
and provides a procedure to better understanding the perceived exchange rate risks for different
stocks across exchanges.
Fall 2014—Honor Thesis--Statistics
II. JOINT MODEL FOR IMPLIED VOLATILITY AND EXCHANGE RATE
Theoretical background1:
In the whole research, for an ADR option, we deal with the foreign stock struck in domestic
currency (here USD)2. The dynamic model of the entire economy, under objective measure P, is
as follows
dX = XαX dt + XσX dW
dSd = Sd αd dt + Sd σddW
dSf = Sfαfdt + SfσfdW
dBd = rdBd dt
dBf = rfBfdt
where
W =
W 1
W 2
W 3
Is a three-dimensional Winner process (with independent components)
For the foreign call struck in domestic currency, the claim, expressed in domestic term, is given
by
Zd = max X T ∗ Sf T − K, 0
Use the Black-Schoels formula we can obtain the price function and the implied volatility is
given by
σf + σX
where σf is the foreign stock volatility and σX is the foreign exchange rate volatility
by no-arbitrage argument, the implied volatility of foreign stock struck in domestic currency
should equal to the implied volatility of domestic options σd
σf + σX = σd
1 Detailed proof refers to Bjoerk book Arbitrage Theory in Continuous Time Chapter 12
2 See Appendix 1 for all the ADRs used in this project
Fall 2014—Honor Thesis--Statistics
if we denote ρ as the correlation between σf and σX , the equation is
σf2 + σX
2 + 2σfσXρ=σd
And we can get the implied exchange rate volatility σX = σf2ρ2 − (σf
2 − σd2) - σfρ
Mean Square minimization3:
In practice, given a stock for a specific day, there are many option choices with different strike
price and expiration date. Notice the implied volatilities have a smile curve. After converting the
strike price of foreign options into domestic price, we can compare the implied volatilities of
foreign options and domestic options for the same stock. Here we use UBS as illustration4.
The data is from Bloomberg on May 28th
.The implied volatility from emerging market is
shown in red and the implied volatility from the ADR market is shown in black. The blue dot
points the current spot price of ADR.
We see the implied volatility differs a lot in some extreme strike price, which is actually not a
problem since in reality those options are not traded in the market. We only need concern about
the implied volatility of options with the strike around the spot price.
3Corroborative work with Professor Tim Leung and Connie Lee in Columbia University in summer 2014
4 See Appendix 2 for more examples and illustrations of implied volatility curves
Fall 2014—Honor Thesis--Statistics
As expected, the curve of implied volatility in foreign market is higher than that in domestic
market (U.S. market), since the ADRs in U.S. market also carry the currency risk that increase
the options price.
We aim to find ρ and σX that minimize the mean square error term between two curves
( ρ , σX )=argmin (σf2 + σX
2 + 2σfσXρ − σd2)i
ni=1
subject to −1 ≤ ρ ≤ 1; σX ≥ 0
where n is total number of points of foreign option strike price struck in USD, σfi is foreign
implied volatility for ith strike price expressed in USD and σdi is the ADR implied volatility of
the same strike price computed by interpolation.
In practice since the strike prices are not exactly the same and we need interpolate either σfi
Fall 2014—Honor Thesis--Statistics
(ADR based methods) or σdi (local based methods) or mix them weighted by the number of
points. Here we examine the implied exchange rate volatility from the implied volatility disparity
(data from 2014/01/02 to 2014/05/29 with expiration date 2014/12/15) of Deutsche Bank (DB in
ADR market and DBK in Germen market). The time series of σX gave us a general impression
of exchange rate dynamics.
Due to the constraint of σX very rarely the optimal σX is 0 and we can just interpret it as the
very low exchange rate volatility. In general the analysis gave us a good impression about the
implied exchange rate volatility, from which we can know people’s expectation about the
exchange rate in advance from the options implied volatility disparity.
The similar analysis of implied correlation between foreign stock prices and exchange rate is
comparatively less significant.
Fall 2014—Honor Thesis--Statistics
Due to the constraint of correlation in the minimization problem there is lots of extreme value
as optimal correlation especially for the ADR based methods. First the extreme value lost its
statistical significance since the optimal value of σX here may be far from correct implied
value without the constraint of correlation. In addition we should look at the nature of implied
correlation and implied exchange rate volatility. People’s expectation could be quite fluctuated in
respond to the information in exchange market, which corresponding the time series ofσX . Yet
the other factor, the implied correlation between the foreign stock and exchange rate represented
people view about the correlation, which is hard to change dramatically during a short time
period. The relative stable implied correlation and fluctuate exchange rate volatility is more
reasonable. Thus to understand the implied exchange rate dynamics we need further exploration
about the other factor, the implied correlation between foreign stock price and exchange rate.
III. SENTIVITY OF IMPLIED EXCHANGE RATE VOLATILITY TO
CORRELATION
The historical correlation between stock price and exchange rate actually is not stable. Here
we take DBK and EUR as illustration
The historical correlation fluctuate from -0.5 to 0.6, which is a very broad range of correlation.
Thus we need to examine how sensitive of to correlation.
To see how the σX is affected by rho I apply the same methods of MSE to DBK and
find the implied volatility of exchange rate but for different rhos obtained from different
time scale .
-0.5000
-0.4000
-0.3000
-0.2000
-0.1000
0.0000
0.1000
0.2000
0.3000
0.4000
0.5000
0.6000
Realized Correlation
Realized Correlation
Fall 2014—Honor Thesis--Statistics
Rho=-
0.18 (1w)
-
0.62
(2w)
-
0.52
(1m)
-
0.55
(2m)
-
0.32
(3m)
-
0.14
(4m)
-
0.06
(6m)
-
0.005
(1y)
0.3
0
(2y)
2014.07.18 13.11 28.01 24.32 25.29 17.56 12.14 10.16 8.91 4.61
2014.10.17 15.74 30.62 26.91 27.88 20.16 14.77 12.77 11.48 6.64
2015.01.16 16.56 31.41 27.71 28.67 20.97 15.59 13.58 12.28 7.31
2016.01.15 18.50 33.88 30.04 31.04 23.07 17.49 15.40 14.03 8.66
𝜎𝑋 from empirical data for last two years:
9.47
The σX is actually quite sensitive to rho so we need choose the proper rho carefully.
One possible solution is to use the rho from same length of maturity (i.e. if we
calculate σX of options with three month maturity, we use the rho from past three
months)
Fall 2014—Honor Thesis--Statistics
Fall 2014—Honor Thesis--Statistics
Fall 2014—Honor Thesis--Statistics
Fall 2014—Honor Thesis--Statistics
IV. DIFFERENT MEASUREMENTS OF CORRELATION
Instead of minimizing the mean square error over σX and ρ we can preset the
correlation and calculate the implied exchange rate volatility. In the original methods,
the implied correlation is extremely unstable and as the analysis above it could
significantly influence the optimal exchange rate volatility. By presetting the correlation,
the minimization problem is simplified to the quadratic minimization
σX =argmin (σf2 + σX
2 + 2σfσXρ − σd2)i
ni=1
Fall 2014—Honor Thesis--Statistics
We can give the explicit formula for σX = σfρ
ni=1
n
From the explicit formula the implied exchange rate volatility is determined by the foreign
implied volatility and correlation between foreign stock and exchange market. The fluctuated
foreign stock and s strong positive correlation between the stock and exchange market
correspond the larger implied exchange rate volatility. By presetting the correlation, the pattern
of implied exchange rate volatility becomes more clear and simple.
The other important parity also came from the explicit formula. The implied exchange rate
volatility represents people’s expectation about the volatility of exchange rate and only depends
on the information on the macroeconomic factors. In other words, the ADRs implication is the
approach to get the implied exchange volatility and theoretically using different ADR stock we
should get the same implied volatility. Back to the formula, which says σX is equal to the
average of product of stock price implied volatility and its correlation with exchange rate for
different stocks. In this project the parity is used to test the result of σX . One potential application
is to compare the implied correlation for different stocks if we know the implied exchange rate
volatility, which may be further studied in the future.
Finally we can use three different measurements of correlation
Correlation as time series over a short period
As discussed previously, we can use the realized correlation as the measurement of
implied correlation since people’s view of correlation is relatively stable.
The realized correlation from 2014.01.02 to 2014.07.18 is within the range between -
0.4 and 0.5. Compared to the implied correlation we got from minimizing both σX and ρ,
the change of realized correlation changes steadily.
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
Realized Correlation (DBK)
Realized Correlation (DBK)
Fall 2014—Honor Thesis--Statistics
Correlation as regression over long period
Another measurement of correlation is the average correlation over long period and
we can regard the correlation as coefficients of regression of change of exchange rate
against stock price. Although compared to the first measurement it’s not dynamic,
instead we can get the confidence interval of the correlation, which will gave us the
range of the implied exchange rate volatility.
lm(formula = diff(DBK[, 2]) ~ diff(DBK[, 1]))
Residuals:
Min 1Q Median 3Q Max
-0.0084750 -0.0018027 -0.0001058 0.0012873 0.0109077
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -1.713e-05 2.347e-04 -0.073 0.942
diff(DBK[, 1]) 8.777e-05 4.956e-04 0.177 0.860
Residual standard error: 0.002961 on 159 degrees of freedom
Multiple R-squared: 0.0001972, Adjusted R-squared: -0.006091
F-statistic: 0.03136 on 1 and 159 DF, p-value: 0.8597
The standard deviation of the slope is pretty small and thus we have a narrow
confidence interval of the correlation as shown in the dashed line. The long-run
Fall 2014—Honor Thesis--Statistics
correlation between stock price and exchange rate is approximate zero with a very tiny
range.
Industrialized Correlation for Different Time Scale
Since the implied exchange rate volatility represent the fluctuation of exchange rate
and should be independent of our choice of specific stocks, we can also looked up for the
aggregate stock market index. Here we can preset the correlation as the correlation
between the bank industrial index (KBW bank index)5 and exchange rate.
From the Comparison of realized correlation of index and realized correlation of a specific
stock (DBK here), as expected, they follow the similar trends over time and the correlation of
index is more stable.
V. CONCLUSION
The exchange market has various forms and is among one of the most volatile and
unpredictable market in the worlds. In addition the unpredicted currency depreciation could
cause serious impact on the whole financial market. For instance, the Brazilian Real lost its value
up to 10% in the recent currency crisis in South America from early 2014. We can find the way
to predict the potential increasing volatility of exchange rate from the implied volatility
5 See Appendix 3 for definition and explanation for KBW Index
-0.5000
-0.4000
-0.3000
-0.2000
-0.1000
0.0000
0.1000
0.2000
0.3000
0.4000
0.5000
0.6000
1
10 19 28 37 46 55 64 73 82 91
10
0
10
9
11
8
12
7
13
6
14
5
15
4
16
3
Realized Correlation (DBK)
Realized Correlation (Index)
Fall 2014—Honor Thesis--Statistics
discrepancy of ADR options in U.S. market and home market. After the financial crisis in 2008,
the U.S. stock market yield a very low return and investors turned to foreign market especially
for emerging market like Brazil for higher return rate. Thus the portion of ADRs increased
significantly and the options market in such countries has rapidly developed, which gave us the
more adequate and statistically significant data. In the project, based on the result of summer
research collaborated with Professor Tim Leung and Connie Lee, I do more sensitivity analysis
for the implied exchange rate volatility respect to correlation between foreign stock market and
exchange rate. By presetting the correlation, the expression for implied exchange rate volatility
becomes simpler and explicit. There are three different measurements of correlation, short term
dynamic correlation, long-term correlation as regression and correlation with industrialized index.
In practice we can combine the measurements and get the more comprehensive impression about
correlation and prediction of implied exchange rate volatility.
The connection between ADR options volatility discrepancy and exchange rate can be further
explored. In the project we choose the bank industrial as example since the banks are more
sensitive to the exchange rate. Since the implied volatility of exchange rate is industry
independent we can possibly examine from other industrial with different sensitivity. All in all,
the methods connected the ADR market and exchange market and provided a good insight of
exchange market dynamics. It’s possible to make further and deeper exploration in the future.
REFERENCE:
[Tomas Bjork, 1998] Arbitrage Theory in Continuous Time chapter 12 Currency Derivatives
p167-181
[Stefan Eichler, Dominik Maltritz, 2008] Currency Crisis Prediction Using ADR Market Data -
An Option-Based Approach, International Journal of Forecasting 26
[Stefan Eichler, Alexander Karmann, Dominik Maltritz, 2009] The ADR Shadow Exchange
Rate as an Early Warning Indicator for Currency Crisis, Journal of Banking and Finance 33
[Dilip B. Madan, 2012] Joint Modeling the Prices of American Depository Receipts, the Local
Stock and the U.S. Dolalr, Journal of Investment Strategy Volume1/Number 4
[Louis Ederington, Wei Guan, 2005] The Information Frown in Option Prices, Journal of
Banking and Finance 29
Fall 2014—Honor Thesis--Statistics
Appendix 1
Name Country Stock
symbol
in US
Stock
symbol
in
original
country
Ratio
DR:ORD
Foreign
currency
Industry
UBS Switzerland UBS UBS 1:1 CHF Bank
Barclay British BCS BARC 1:4 GBX Bank
Santander Spain SAN SAN 1:1 EUR Bank
GlaxoSmithKline British GSK GSK 1:2 GBX healthcare
Novartis Switzerland NVS NOVN 1:1 CHF healthcare
TEVA Israel TEVA TEVA 1:1 NIS healthcare
British
Petroleum
British BP BP 1:6 GBX Oil and gas
Royal Dutch
Shell
British RDS RDS 1:2 GBX Oil and gas
Alcatel-Lucent France ALU ALU 1:1 EUR telecommunication
America Movil Mexico AML AML 1:20 MXN telecommunication
China Mobile China CHL 941 1:5 HKD telecommunication
Cemex Mexico CM CEMEX 1:10 MXN Building matrials
BHP Biliton Australia BHP BHP 1:2 AUD Metal and mining
Nokia Finland NOK NOK1V 1:1 EUR Tech equipment
Fall 2014—Honor Thesis--Statistics
Appendix 2
In the plots, the ADR volatility are shown in black and the original market volatility are shown in
red. All the strike price is converted into USD using the spot exchange rate assuming the
exchange rate doesn’t change until expiry. The blue doc points out the spot price and the blue
dashed lines are 75% and 125% of spot rate. Basically we only need compare the implied
volatility between the dashed lines.
Fall 2014—Honor Thesis--Statistics
Fall 2014—Honor Thesis--Statistics
Fall 2014—Honor Thesis--Statistics
Appendix 3
The KBW bank index is an economic index consisting of the stocks of 24 banking companies.
This index serves as a benchmark of the banking sector. This index trades on the Philadelphia
Stock Exchange, where it was created. The KBW Index is named after Keefe, Bruyette and
Woods, a recognized authority in the banking industry. The KBW Index trades under ticker
symbol BKX. The index is weighted according to capitalization and represents major banks and
money centers from across the country. Mathematically, the index is based on a tenth of the
value of the Keefe, Bruyette and Woods Index (KBWI). It began trading options in September of
1992.